Short-Time Dynamics Reveals Tg Suppression in Simulated

Jul 14, 2017 - Suppression of the glass transition temperature, Tg, in polymer thin films is of great practical importance and theoretical significanc...
2 downloads 13 Views 4MB Size
Download PDF
Article pubs.acs.org/Macromolecules

Short-Time Dynamics Reveals Tg Suppression in Simulated Polystyrene Thin Films Yuxing Zhou and Scott T. Milner* Department of Chemical Engineering, Pennsylvania State University, University Park, Pennsylvania 16802, United States S Supporting Information *

ABSTRACT: Suppression of the glass transition temperature, Tg, in polymer thin films is of great practical importance and theoretical significance. It is widely believed that such Tg suppression results from enhanced segmental mobility at the free surface. To investigate these effects, we carry out unitedatom molecular dynamics simulations on free-standing polystyrene thin films. Care has been taken to ensure consistent behavior between thin films and the bulk. The dilatometric Tg inferred from the density versus temperature shows substantial reduction in thin films compared to the bulk even at high cooling rates. Furthermore, we find that dynamical Tg shifts, obtained by collapsing temperature-dependent shorttime dynamical properties onto a master curve, vary with film thickness just like the dilatometric Tg. We apply the same data collapse procedure to dynamics of segments within a given distance from the free surface to obtain the local Tg(z), which reveals a mobile surface layer of about 4 nm.

■

Forrest showed that for a 6 nm thick polystyrene film supported by Pt coated SiN substrate no appreciable suppression of Tg can be observed for cooling rates larger than 90 K/min (or 0.6 Hz).5 Similar cooling rate dependence of thin film Tg has been observed in DSC measurements.6 These results appear to clarify some contradictions in the literature, such that thickness dependence of thin film Tg is weak or even absent in scanning calorimetric measurements with fast cooling rates7 or AC calorimetric8 and dielectric spectroscopy2,9 measurements at high frequency. Investigations of dynamics near free surfaces also indicate that enhanced mobility is only observable at very long time scales or low frequencies.10 In contrast to the extensive experimental literature, there are rather few atomistic MD simulations of the glass transition in polymer thin films, largely because it is computationally expensive to simulate systems with slow dynamics. Since the longest time accessible in MD simulations (∼1 μs) is orders of magnitude smaller than in experiments, one would expect no thickness dependence of Tg according to Fakhraai and Forrest. Indeed, recent atomistic simulations on supported PS films11 with a cooling rate of 10 K/ns find the average Tg determined from the temperature dependence of density is almost identical to the bulk value for films as thin as 2 nm. On the other hand, enhanced mobility near a free surface is a common feature in coarse-grained simulations,12−15 atomistic MD simulations,11,16,17 and lattice models18 of thin films with a

INTRODUCTION While the of nature of the glass and the glass transition in bulk materials remains a subject of intense debate, nanoconfinement effects on glass-forming polymer thin films have attracted a great deal of interest in recent years. This interest is largely driven by the importance of glassy thin films in engineering applications ranging from protective coatings to organic photovoltaics as well as fundamental questions of glassy dynamics such as the origin of non-Arrhenius relaxation behavior, cooperative rearrangements, and dynamical heterogeneity. Since the first observation of the confinement effect on glass transition temperature Tg in supported polystyrene thin films by Keddie et al.,1 a variety of experiments have shown considerable deviation of Tg in polymer films from the bulk value, when the film thickness is below some tens of nanometers. The average Tg for supported thin films has been reported either increasing, decreasing, or unchanged compared to the bulk depending on polymer chemistry, polymer−substrate interaction, annealing history, and measurement techniques.2 In contrast, there is a growing consensus that Tg is lower in free-standing films as a result of highly mobile layers near the free surfaces, although some experimentalists dispute this.3 (One exception is reported for star-shaped polymer thin films, for which the Tg of free surface layer can be slightly higher than the bulk probably because of higher packing densities of short-arm star-shaped molecules at the external interfaces.4) Another important experimental finding is that the Tg reduction in thin films depends strongly on the cooling rate. On the basis of ellipsometric measurements, Fakhraai and © XXXX American Chemical Society

Received: May 4, 2017 Revised: July 3, 2017

A

DOI: 10.1021/acs.macromol.7b00921 Macromolecules XXXX, XXX, XXX−XXX

Article

Macromolecules

master curve approach. A mobile surface layer of about 4 nm with lower Tg is observed, which agrees with the thickness of the mobile surface layer inferred from long-time relaxation in our simulations as well as some experimental measurements.27,28 Our results suggest that while Tg is most commonly associated with the relatively slow dynamics of α relaxation, suppression of Tg in thin films can also be revealed by shorttime dynamics.

free surface. Results from these simulations often predict appreciable suppression of average or local Tg, by extrapolating the segmental relaxation time to experimental time scale (e.g., 100 s) using VFT-like equations. It is unclear whether the disagreements in measured and simulated Tg reduction arise from probing different relaxation processes (such as dipole moment relaxation, segmental mobility, and density fluctuations) that are coupled with different characteristic time scales, since the robust correlation for bulk systems between Tg and many dynamical processes may no longer hold for thin films.10,19 In addition to the slow α relaxation, which is thought to involve collective motions by more than one particle, supercooled liquids also exhibit a fast β relaxation process, which is thought to be related to rattling motions of particles within their cages formed by neighboring particles and often manifests as the plateau in the time-dependent mean-square displacement. From experimental observations that T g reduction in thin films is not appreciable at short time scales, one may expect the effect of a free surface on the β relaxation will be weak or even absent. However, contradictory and somewhat confusing results have been reported. For example, measurements of structural relaxation in supported PMMA films20 suggest that the free surface effect on the structural relaxation rate, which is possibly related to the β relaxation process, is stronger than on the local Tg associated with α relaxation. In addition, simulations of free-standing PS films show that while the average β relaxation rate (based on orientational relaxation time of phenyl bonds) is the same for films and the bulk, the β process is faster in the center of the film than near the free surface, as opposed to the α relaxation.17 Furthermore, previous works studying the Debye−Waller factor, a measure of vibrational motion within the β-relaxation, in general find a good agreement between short-time meansquare displacement ⟨u2⟩ and α-relaxation time for bulk materials,21,22 even at the particle level.23 However, for supported thin films the Tg shift based on Debye−Waller factor from incoherent neutron scattering has been reported to be inconsistent with the Tg shift conventionally obtained from thermodynamical properties.24−26 In simulations, most studies have focused on bead−spring models, even though chemistryspecific attributes at short-time scales may play an crucial role in the polymer glass transition.15 To summarize: despite many recent experimental and simulation studies, the effects of polymer interface(s) on different relaxation processes or time scales, the origin of the discrepancy in Tg shifts from different measurements, and the extensibility of coarse-grain models are still not fully elucidated. In this work, we perform united-atom molecular dynamics simulations on free-standing polystyrene films and investigate the effects of a free surface on thin film Tg, inferred from a variety of film-average and local properties. The dilatometric Tg inferred from film density versus temperature shows a noticeable thickness dependence, with a reduction of about 35 K for a 6 nm thick filmin contrast to the expectation from experiment of no Tg reduction at high cooling rates. For the same cooling rates, we also obtain Tg values by collapsing the temperature dependence of short-time dynamical properties for different films and bulk onto a master curve. The Tg values obtained this way are generally consistent with the dilatometric Tg. We also extract the local Tg from relaxation of monomers and chains at a given distance from the free surface, using a similar

■

METHODS

We perform molecular dynamics simulations on a united-atom model of atactic polystyrene, with each chain consisting of 10 monomers (Mw = 1040 g/mol). These chains are quite short compared to most experimental systems. We use short chains to keep the simulation time manageable. We have checked for chain length dependence by simulating 20-mer systems for a few selected film thicknesses, which exhibit thickness-dependent Tg shifts consistent with our 10-mer simulations. Each thin film system contains 384, 300, 192, 140, 110, or 80 chains, corresponding to a film thickness of approximately 28, 22, 14, 10, 8, or 6 nm at 500 K with box area fixed to 5 × 5 nm2. The atactic chains are constructed by duplicating a single isotactic chain, turning off the improper dihedral potential that stabilizes the tetrahedral backbone carbon and equilibrating at high temperatures to randomize the tacticity, and then restoring the dihedral potential. For free-standing films, periodic boundary conditions are applied only in the x and y directions, while an open boundary is used in the z direction. Bulk systems consisting of 192 chains are also prepared with full periodic boundary conditions in all directions. To improve statistics, results are averaged over 5−20 independent runs depending on system size. Thin film systems are equilibrated in the NVT ensemble at 500 K until monomers have diffused farther than the average end-to-end distance and then quenched from 500 to 200 K at a cooling rate of 5 K/ns. Bulk systems are equilibrated and quenched in the same way, except that the semi-isotropic NPT ensemble is used, with varying system dimension and zero pressure in the z direction only. The bulk system dimensions are fixed in x and y for consistency with the thin films, for which fixed x and y dimensions are necessary to keep the free surface from shrinking in response to surface tension. Compared to cooling results from full 3d NPT simulations, we find the bulk densities are identical in the melt region; in the glassy state, the density is slightly higher in 3d NPT systems, since the system can contract laterally in response to the tensile stress caused by cooling, which the system with fixed transverse dimensions cannot do. All simulations are carried out using the GROMACS package29 with integration time step of 2 fs. Stochastic velocity rescaling thermostat (τt = 0.2 ps) and the Berendsen barostat (τp = 0.5 ps) are chosen for NVT and NPT ensembles. The united-atom force field we use for PS is adapted from the TraPPE-UA model, which has been shown to produce polymeric properties in good agreement with experimental values for simple polymer systems (e.g., PE30 and iPP31). However, it is found that for PS the UA model predicts a much faster dynamics (∼40 times larger) compared to the all-atom (AA) PS model,32 probably because of a relatively weak dihedral potential along the aliphatic backbone in the presence of phenyl ring.33 To remedy this, we reparametrize the dihedral potential along the aliphatic backbone, as well as the improper dihedral used to maintain the sp3 stereochemical configuration of the CH group, to match the corresponding dihedral distributions obtained from the AA model, using the standard Boltzmann inversion method (see Appendix B for the modified force field). The resulting modified UA PS model leads to about 10 times faster diffusivity than the AA model for all temperatureswhich is fortunate, since the AA model dynamics is about 4.6 times slower than experiments.34 In addition, the modified UA force field gives a slightly smaller bulk density than the original TraPPE UA model, only about 4% higher than the experimental values for PS with Mw ∼ 910 g/mol at 0 MPa between 400 and 500 K. This good agreement for the density contrasts with the relatively large deviations given by the AA model (density 7% B

DOI: 10.1021/acs.macromol.7b00921 Macromolecules XXXX, XXX, XXX−XXX

Article

Macromolecules too high) as well as another widely used UA model (18% too high) proposed by Lyulin et al.35,36 As a result, we do not need to boost the external pressure to match the experimental density as in previous works,11,35 which anyway is not possible for simulating films exposed to vacuum. Usually, simulations use cutoff LJ interactions and apply standard “dispersion corrections” to add back the average attractive energy of distant monomers. These corrections contribute to the cohesive energy and increase the density at a given pressure. (We included the standard dispersion correction for the benchmarking bulk simulations discussed above.) This approach works well for homogeneous 3d periodic systems. However, for systems with interfaces, computing these corrections is more involved. One must consider the actual configuration during the MD run and use some special treatments.37,38 Unfortunately, this nonuniform long-range correction is not currently implemented in the GROMACS package. To make a consistent comparison between the bulk and thin films, we turn off the dispersion correction and use a rather large cutoff distance of 1.8 nm for both bulk systems and thin films. While the bulk density without the long-range correction decreases by less than 1%, a consistent density value is now reached in the center of the film for all film thickness studied. We note that the density match is crucial for a proper comparison of dynamics between thin films and the bulk. For example, in a recent atomistic simulation study of supported PS films,16 the segmental dynamics of the middle layers of a rather thick film (14 nm) is noticeably faster than the corresponding bulk system. In contrast, we find in our simulations with the dispersion correction turned off that the segmental relaxation time of the middle layer in a 14 nm film is the same as the bulk. Another issue we have encountered is that a free-standing film tends to develop spurious shear modes along the z direction, which are more pronounced in thick films near the free surfaces. This is probably due to the lack of sufficient friction in the xy plane, as the transverse area is anomalously small compared to the film thickness to keep the overall system volume manageable (see Appendix A for details). A closer examination reveals that particles in each layer tend to “drift” together, with their center of mass moving like a random walk. This shearlike motion cannot be corrected by removal of the center-of-mass motion of the entire system. To minimize this finite-size effect without increasing the box area, we subtract the lateral center of mass motion in each layer before analyzing the local dynamics. For bulk systems, this shearlike motion is largely inhibited by the full periodic boundary conditions in a cubical box. However, small fluctuations of the center of mass do exist in each layer, which scales approximately with the number of particles in the layer N as √N. To make a consistent comparison between thin films and bulk systems, we apply the same corrections to bulk samples as for thin films. In practice, we divide the system into a number of layers of thickness about 0.7 nm, which is thin enough to subtract the average shearlike motion in each layer while still thick enough to containing plenty of particles for good statistics. We note that this artifact does not affect the measurement of bond rotational dynamics, given that the zgradients of the shear mode is small compared to the size of a bond.

width of the interface is about 1 nm, which is expected to decrease at lower temperature.

Figure 1. Snapshots of the bulk and thin films with selected thicknesses at 500 K (top) and the corresponding density profiles as a function of z position (bottom). Solid curves are fits to eq 1. Film thickness h is defined as the distance between two free surfaces indicated by dashed lines (inflection point of density profile).

We obtain the average density ρ̅ by averaging the density profile between the two free surfaces, defined by the inflection points of the profile (dashed lines in Figure 1). Also shown in Figure 1 is the bulk density at 500 K, which agrees with the plateau density ρb of thin films. (This agreement depends on turning off the long-range corrections in the bulk simulations, as discussed above.) It is also interesting to define an interior density ρ0 that characterizes the middle region of the film. Naively, ρ0 should be the same as the plateau density ρb from the fit. However, the error function profile may not fit well for nonequilibrium systems as a result of faster aging near the free surface versus the middle of the film.39 To avoid this potential problem, we define the interior density ρ0 as the average over the middle region in the thin film 1.5 nm away from each free surface, which is expected to give the bulk density for systems in equilibrium, given that the width of free surface is about 1 nm. We perform the density calculation for each configuration collected every 40 ps during the cooling procedure. As shown in Figure 2a, the average density for thin films is smaller than the bulk and decreases with decreasing film thickness. This is simply a result of the contribution of the less-dense interfacial region, which is a larger fraction of a thinner film.

■

RESULTS Tg from Temperature Dependence of Density. One way the glass transition reveals itself is a relatively sharp decrease in the thermal expansivity as the system solidifies. To observe this in simulation, we begin by obtaining the average density of a thin film, evaluated between two polymer−vacuum interfaces. To locate the interfaces, we fit the film density profile to an error function ρ(z) = ρb /2(1 + erf((z − z 0)/σ ))

(1)

where z0 is the position of an interface, ρb is the plateau density well inside the film, and σ characterizes the interfacial width. This expression well describes the density profiles, as shown in Figure 1 for selected film thicknesses equilibrated at 500 K. The C

DOI: 10.1021/acs.macromol.7b00921 Macromolecules XXXX, XXX, XXX−XXX

Article

Macromolecules

Figure 2. (a) Average density ρ̅ and (b) interior density ρ0 as a function of temperature for different systems. The solid are nonlinear fits to eq 2.

dependence of Tg from the density versus temperature in simulations, in the same way as for typical dilatometric measurements. Figure 3 shows dilatometric Tg values obtained as a function of film thickness, from analyzing both the average density and

In contrast, the interior densities are essentially the same for all films and the bulk in the liquid regime (T > 400 K) above the glass transition. In the glassy regime, the interior density of the thickest film is very close the bulk but increases systematically with decreasing film thickness (see Figure 2b). It appears that the interior of a thin film ages faster and thus becomes more dense because of the nearby free surface. This observation suggests nonlocal effects of the free surface on dynamics that extends more than 1.5 nm into the interior. In principle, we could probe the dynamical length scale over which the free surface accelerates aging within the film by the depth to which the nonequilibrium local density is perturbed by the free surface. However, large spatial fluctuations and insufficient sampling of local density at low temperatures make accurate measurement during the cooling process difficult. Instead, we investigate the length scale of enhanced dynamics directly from short-time local dynamical properties, as discussed in the following section. We determine the Tg values of thin films and the bulk by fitting the density versus temperature to a function that smoothly transitions from a constant thermal expansivity in the glass to a higher value in the liquid. We use the following empirical form40

Figure 3. Film thickness dependence of Tg inferred from the change in the density. Different parts of thin films are considered, i.e., 0.0 nm (average density), 1.5 nm (interior density), and 3.0 nm (core density), from the two free surfaces. The solid lines are fits of data to ΔTg(h) ∼ 1/h, and the dashed line corresponds to the bulk value. The error bars are calculated based on the standard errors of corresponding fits weighted by measurement errors.

⎛ T − Tg ⎞⎤ ⎛ M − G⎞ ⎡ ⎛ M + G⎞ ⎟ ln⎢cosh⎜ ⎟ + c ρ(T ) = w⎜ ⎟⎥ + (T − Tg)⎜ ⎝ 2 ⎠ ⎣ ⎝ 2 ⎠ ⎝ w ⎠⎦

(2)

the interior density of the films. We find the dilatometric Tg from average density decreases substantially with decreasing film thickness, varying roughly as ΔTg ∼ 1/h. This 1/h dependence reflects the decreasing contribution to the average density of a mobile surface layer with low Tg and roughly fixed thickness to an increasingly thick film with essentially bulklike properties. In contrast, the Tg reduction of the interior region is weaker than that of the whole film, and only substantially reduced for thinner films, because the outermost 1.5 nm of material has been eliminated from the average. Evidently some near-surface material deeper than 1.5 nm is still more mobile, with a lower Tg than the bulk value, so that the dilatometric Tg from the interior density still shows a modest decrease for the thinnest films. Qualitatively, this behavior is consistent with the relatively weak thickness dependence of the interior density versus temperature of Figure 2b. By eliminating more surface material, e.g., the outermost 3.0 nm from each free surface, we find the Tg of the middle layer becomes essentially the same as the bulk Tg within the error bars, independent of the film

where w is the width of the transition, M and G are the slopes of melt and glass regions, and c is the density at Tg. If we fit the data with w as free parameter, its value ranges from 40 to 100 K for different films. This is a much broader glass transition than seen in experiments, where w ranges from 2 to 5 K.40,41 This increased transition width may be due to smaller fragility at the relatively high Tg probed in simulations with high cooling rate. In practice, we find that fixing the transition width w = 80 K yields good fitting results for thin films and the bulk, as shown in Figure 2. We emphasize that the Tg value is rather sensitive to the temperature range over which the fit is performed. It is important to ensure that fully developed glassy and melt regions are included in the fit. For example, fitting only the data from 300 to 500 K for the thinnest films, where the system still remains partially liquidlike, would lead to a narrower transition, larger glassy thermal expansivity and hence higher Tg value. Moreover, in that temperature range, different choices of w may vary Tg considerably without significantly changing the fit quality. With these caveats in mind, we can obtain the thickness D

DOI: 10.1021/acs.macromol.7b00921 Macromolecules XXXX, XXX, XXX−XXX

Article

Macromolecules thickness (no middle layer for the 6 nm film), as shown in Figure 3. In short, we observe a dilatometric Tg reduction of about 20 and 36 K, respectively, for the interior region and the whole film of 6 nm. This is in contrast to previous simulation results11 and experimental predictions5 for supported films on a neutral substrate that no appreciable Tg reduction can be observed at high cooling rates. Even if we take into account that the Tg reduction in a supported film is about half as large as that of a free-standing film of same thickness, a Tg reduction of about 10−18 K in a 6 nm supported PS film is still significantly larger than one would expect based on the experimental value of 10 K for the same film thickness but with a cooling rate 12 decades slower. On the other hand, we note that the magnitude of Tg reduction is smaller than the experimental values for freestanding PS films obtained from ellipsometry and Brillouin light scattering (BLS) techniques,42 e.g., 70 K for a 20 nm film. It is possible that this discrepancy is due to the different time scales probed in simulation (t ∼ 0.1 ns) and in experiments (t ∼ 1 s), such that the cooperative α relaxation at long times may have additional contribution to Tg reduction in thin films aside from the local short-time β relaxation. Interestingly, the thickness dependence of the Tg reduction appears much closer to that determined from direct measurement of probe reorientation in PS films,27 e.g., 28 K for a 14 nm film. We emphasize that our modified force field can reproduce the correct bulk Tg rather quantitatively, suggesting that our united atom simulations are a reasonable representation of real PS. We note that our simulated bulk glass transition = 402 K is comparable to previous simulation temperature Tbulk g results for similar PS systems11,17and much higher than the nominal experimental Tg of 374 K. But to compare our Tg with experiment, we must account for the effects of our very short chains and very high cooling rate. On the one hand, Tg for low Mw = 103 PS is reduced by about 88 K, estimated from the extrapolation of experimental data using the Fox−Flory equation,43 Tg(Mn) = T∞ g − K/Mn, with K = 8.8 × 104 K g/mol. On the other hand, the fast cooling rate γ = 5 × 109 K/s in the simulation leads to an increase of Tg by approximately 108 K, compared to a typical experimental cooling rate of about 1 K/s, according to the Vogel−Fulcher−Tammann (VFT) equation,44 log γ = A − B/ (Tg − T0), with A = 13.5, B = 570 K, and T0 = 333 K. Given the typical experimental Tg of about 374 K for high Mw PS at small cooling rates,41,43−45 the expected Tg at simulation conditions is estimated as 394 K, in reasonable agreement with the simulation result of 402 K. Such corrections for both cooling rate and molecular weight effects are not yet routinely applied in simulations of glassy polymers but have been previously discussed for atomistic simulations.46−48 The cooling rate dependence is typically accounted for in bead−spring chain models by VFT-based extrapolation.12,13,49 Tg from Short-Time Dynamics. To explore the connection between Tg and dynamics in thin films, we measure different short-time dynamical properties for the same cooling runs and find the shifts Tg from dynamical scaling. In brief, we find that a given short-time dynamical property versus temperature for different films and bulk samples collapses onto a master curve, after shifting the temperatures by an amount that depends on film thickness. From the temperature shifts we determine the suppression of Tg relative to the bulk as

displayed in Figure 4, which shows similar thickness dependence as the dilatometric Tg obtained above.

Figure 4. Film thickness dependence of ΔTg obtained from data collapse of different short-time dynamical properties: monomer and chain center-of-mass mean-square displacement (MSD) and backbone mean-square angular displacement (MSAD). The solid lines are fits to ΔTg(h) ∼ 1/h. Error bars are smaller than the symbols.

Now we describe in more detail what short-time dynamical properties we use to extract Tg shifts versus film thickness, how those data are analyzed, and how the master curves are constructed. The dynamical property we examine first is the lateral monomer mean-square displacement (MSD), i.e., how far monomers can diffuse laterally in a given short time. As mentioned above, the effect of spurious shear modes must be eliminated before measuring the monomer MSD. In Figure 5,

Figure 5. Lateral mean-square displacement of monomer center of mass as a function of time for a 14 nm thick film (black lines) and bulk (red lines), at 500 and 400 K, before (dashed lines) and after (solid lines) removing the spurious shear mode described in the main text. The dashed lines represent the end-to-end distance Ree ≈ 16.7 Å and the average diameter of united atoms σ ≈ 4.3 Å, which is also close to a monomer size.

we show both raw and corrected monomer MSD as a function of time for a 14 nm thick film and the bulk, at T = 500 and 400 K. We emphasize that although the effect of the shear modes on average MSD appears to be modest, it can significantly alter the apparent local dynamics in films (see Appendix A for details). To extract a distance from observations of monomer MSD, we must choose an appropriate observation time t*. Three dynamical regimes for an unentangled polymer melt can be identified for the monomer MSD: the initial ballistic regime with ⟨r2(t)⟩ ∼ t2, an intermediate subdiffusive regime with ⟨r2(t)⟩ ∼ t0.66 corresponding to the Rouse-like motion of E

DOI: 10.1021/acs.macromol.7b00921 Macromolecules XXXX, XXX, XXX−XXX

Article

Macromolecules

Figure 6. Temperature dependence of average dynamical properties at t* = 200 ps for bulk and films of different thicknesses. (a) Mean-square displacement of monomer center of mass. (b) Mean-square angular displacement of CH2−CH2 backbones. (c) Mean-square displacement of chain center of mass. Insets: master curve of corresponding quantity as a function of temperature shifted by ΔTg(h), in which the cyan curve is polynomial fit g(T) for the bulk data.

collapse procedure to obtain ΔTg. Note that the monomer MSD is always in the subdiffusive regime for this range of time at T ≤ 500 K, but monomers do become diffusive at higher T. We find that the data collapse is equally good for all t* values, although the master curves do have different shapes for different t*. The ΔTg values for different film thicknesses depend only weakly on t*, as shown in Figure 7.

monomers within a chain (similar exponent is observed for bead−spring chains50), and the long-time diffusive regime with ⟨r2(t)⟩ ∼ t as monomers diffuse farther than the chain end-toend distance. We choose an observation time t* of 200 ps, which corresponds to monomer displacements of less than one monomer diameter, within the subdiffusive regime of chain dynamics over the temperature range of interest (see Figure 5). The mean-squared displacement of monomers ⟨r2(t*)⟩ is calculated for different starting times or, equivalently, different temperatures, along the cooling trajectory. The data are finally averaged over a short time interval of 1 ns to improve the statistics. We emphasize that our results for Tg shifts are insensitive to the choice of observation time t*, over a range of short times from 100 to 400 ps, as discussed below. Previous simulation studies17,48,51 have estimated Tg from the intersection of the melt- and glassy-state temperature dependences of short-time MSD. (This approach is similar to the way dilatometric Tg is obtained from the intersection of asymptotic tangents to the density versus temperature curve.) However, it is difficult to locate precisely the “kink” in the temperature dependence of monomer MSD, as the transition between melt and glassy state is quite smooth (see Figure 6a). Instead, we find that the temperature dependences of ⟨r2(t*)⟩ for both thin films and the bulk follow the same functional form upon shifting the temperature by a thicknessdependent factor ΔTg(h), which is directly related to Tg reduction. To construct the master curve in a systematic way, we first fit the T-dependent ⟨r2(t*)⟩ for the bulk system to a sixth-order polynomial g(T), as a convenient form that fits the bulk data, and then obtain the shift factor ΔTg for each film thickness by fitting the T-dependent data to g(T + ΔTg). Since the polynomial fit becomes significantly worse for T < 200 K or T > 500 K, we fit the thin film data lying within the temperature range so that 200 K < T + ΔTg < 500 K by iteration. As shown in the inset of Figure 6a, the data collapse is excellent for all thin films and the bulk system. The resulting ΔTg(h) varies strongly with film thickness as shown in Figure 4. These results agree quantitatively with the Tg reduction inferred from the average density versus temperature (see Figure 3). We note that this master curve construction method is more robust in obtaining Tg reduction, as compared to fitting the temperature dependence of density, which is sensitive to the fitting procedure as discussed previously. To investigate effect of the observation time t* on our results, we vary t* from 40 to 400 ps and perform the same data

Figure 7. Average ΔTg obtained from data collapse of monomer MSD observed at different time t* for various film thicknesses.

This insensitivity of ΔTg to the choice of t* not only demonstrates that our method is robust for obtaining Tg in thin films but suggests that a film of thickness h at temperature T looks dynamically the same as the bulk at higher temperature T + ΔTg(h), at least in the subdiffusive region. Our observations are consistent with the idea of “rheological temperature”the temperature at which the bulk material would have the same local dynamical propertiesrecently proposed to describe the depth-dependent dynamics in thin films.52 Indeed, bead−spring chain simulations of supported50 and free-standing films39 reveal similar dynamical behavior for films and the bulk at the same reduced temperature, i.e., intermediate-time monomer MSD versus T − Tc, where Tc is the mode-coupling theory (MCT) glass transition temperature obtained from a power law fit of relaxation time and the aging rate versus T − Tg. Monomer MSD is not the only dynamical property we can use to extract Tg shifts from short-time dynamics. We perform the same data collapse method for other dynamical properties, including bond orientation dynamics and chain center-of-mass motion, and obtain corresponding ΔTg values. In Figures 6b and 6c, we show the original data and the master curves for mean-square angular displacement (MSAD) of backbones and the MSD of chain center of mass for t* = 200 ps. The data F

DOI: 10.1021/acs.macromol.7b00921 Macromolecules XXXX, XXX, XXX−XXX

Article

Macromolecules

Figure 8. Same as Figure 6 except for local dynamical properties (t* = 200 ps) at different distances d from the free surface in a 14 nm film. Insets: master curves of each dynamical property as a function of temperature shifted by ΔTg(d).

collapse is very good for both quantities. The angular displacement for a given backbone vector vi⃗ , which connects two consecutive aliphatic CH2 atoms separated by one CH atom, is calculated as Δθi(t) = arccos(vi⃗ (0)·vi⃗ (t)/|vi⃗ (0))|2), By definition, Δθi is bounded to the maximum value of π at long times. However, within the observation time t* = 200 ps, we observe no bond rotates more than 180° at the highest temperature 500 K. The Tg reduction obtained from backbone rotational motions (mean-square angular displacement) and chain center-of-mass translational motions (mean-square displacement) as a function of film thickness also are shown in Figure 4. In comparison to the ΔTg previously obtained from monomer MSD, they agree with each other and also with the dilatometric Tg, which further suggests that substantial Tg reduction can be revealed by short-time measurements. We notice that the suppression is relatively weaker for the Tg determined from bond orientational motions than from monomer MSD, while the chain translational motions show stronger Tg suppression. One may ask whether the difference in the average Tg reduction for different dynamical properties arises from their distinct responses to the free surface or from a differently weighted average of essentially the same local response. To this end, we measure the local dynamics within a film as a function of the distance from the free surface and apply the same data collapse approach to obtain the local Tg reduction. We divide the system into several “reporting layers” about 0.7 nm thick, each containing a subset of the centers of mass of the monomers, backbones, or chains. Objects in each layer are selected based on their average position during t* = 200 ps, which is sufficiently small so that monomers diffuse less than the layer thickness even at the highest temperature of 500 K. (Note that the position of the outermost “surface” layer defined this way will not coincide exactly with the PS−vacuum interface defined by inflection point in the density profile.) As depicted in Figure 8, the layer-resolved monomer MSD, backbone MSAD, and chain MSD are enhanced near the free surface compared to the bulk values. Overall, we find a good collapse of each dynamical property for different layers in the film of 14 nm. (For the outermost layer about 0.5 nm from the free surface, the low temperature dynamics slightly deviates from the master curve, probably because of a smaller density than the inner layers (see Figure 1), so the quantitative results for this layer should be interpreted with caution.) Figure 9 shows results for the local Tg shift versus distance from the free surface d, obtained from data collapse of local short-time dynamics of the monomer MSD, backbone MASD, and chain MSD. Remarkably, the results are essentially identical

Figure 9. Suppression of local Tg as a function of the distance from the free surface d obtained from data collapse of different dynamical properties. The profile has been symmetrized over two halves of the film. Solid curve is a guide to the eye.

for the different dynamical properties, in contrast to the average Tg reduction shown in Figure 4. We argue that the local Tg reduction might be more fundamental than the average Tg reduction, and the seemingly inconsistent results can be reconciled as follows. Given the successful collapse of local dynamics, the Tdependence of some local dynamical property can be described as ϕi = g(T + ΔTi) for each layer i, with g(T) the functional form for the master curve. The average dynamics is then obtained by averaging over the local dynamics weighed by the mass of each layer ϕ = ⟨ϕi⟩ = ⟨g(T + ΔTi)⟩ ≈ g(T + ΔT), where the last approximation is suggested by the successful data collapse of average dynamics. It is thus clear that while the local Tg reduction is independent of the type of dynamical measurement, the data collapse of an average dynamical property might only be a good approximation so that the average Tg reduction obtained from the shift factor ΔT depends on the specific form of g(T). In fact, the effect of different weightings over local Tg gradient has been examined to explain the discrepancy between different measurements of the mean film Tg.53−55 The local Tg profile of Figure 9 reveals a mobile surface layer of about 4 nm thick, beyond which bulklike behavior is recovered in the middle of the film. This nonlocal effect of confinement on dynamics is consistent with the previous observation that the interior density can be affected by the free surface much farther away than the distance (∼1 nm) at which the bulk density is recovered in an equilibrium system. As an aside, we find that results based on rotational dynamics of side groups, characterized by the vector between the C atom G

DOI: 10.1021/acs.macromol.7b00921 Macromolecules XXXX, XXX, XXX−XXX

Article

Macromolecules

Figure 10. (a) Layer-resolved autocorrelation function of P2 for a 14 nm thin film at 400 K. The solid lines are to the KWW equation. (b) Local segmental relaxation time τ versus distance from the free surface at different temperatures: 500, 480, 460, 440, 430, 420, 410, and 400 K (from lower to upper). Solid curves are fits to an error function; dashed lines are bulk values. The profile has been symmetrized over two halves of the film. (c) Bulk relaxation time τ/τ0 versus 1/T and (ξ/T), normalized by values at T0 = 500 K. Solid line is a VFT fit while dashed line is a linear fit. The inset displays free surface size versus temperature. (d) Depth dependence of local Tg, obtained from extrapolation of segmental relaxation time to 100 s. Same solid curve as in Figure 9 is shown for comparison. Inset: local fragility m as a function of d with bulk value indicated by dashed line.

attached to the backbone and the para-CH atom on the phenyl ring, are quantitatively similar to those based on the rotations of backbone tangents. To this point, all our local Tg values, and the inferred range of influence of the free surface, have been based on short-time dynamical properties. Alternatively, we can investigate the range of influence of the free surface by analyzing the segmental autocorrelation function decay time, typically of order 1−1000 ns. We cannot extract isothermal autocorrelation functions at such long times from our rapid cooling scans. Instead, we fully equilibrate 14 nm films and bulk systems at different temperatures ranging from 500 to 400 K and then calculate the segmental relaxation time τ. Here τ is defined as the time when the autocorrelation function of the second Legendre polynomial P2(t) = (3/2)⟨cos2 θ(t)⟩ − 1/2 decays to 1/e, where θ(t) is the backbone angular displacement defined above. Figure 10a displays an example of the autocorrelation function of local P2(t) at different distances from the free surface for a 14 nm film at 400 K. These autocorrelation functions are well fit by the Kohlrausch−Williams−Watts (KWW) stretched exponential function, P2(t) = A exp[−(t/ τK)β]. We find the stretching exponent β varies from a lower value (∼0.5) near the free surface to the bulk value (∼0.7) in the center regardless of the film thickness (see Supporting Information). We then obtain the local segmental relaxation time as a function of position in the film for each temperature, as shown in Figure 10b. In the temperature range studied, the extent to which the long-time dynamics is perturbed by the free surface (3−5 nm) is similar to the range over which the local Tg obtained from short-time dynamics is perturbed. This 3−5 nm length scale is also consistent with the experimental value of 4− 5 nm at 370 K deduced from probe reorientation27 and nanoparticle embedding measurements.28

Moreover, we notice a moderate increase of the size of the mobile surface layer with decreasing T. According to the thermodynamic framework of glass transition, such as Adam− Gibbs (AG) theory56 and random first-order transition (RFOT) theory,57,58 the growing relaxation time below the activation temperature can be related to some growing correlation length ξ as τ ∼ exp(ξψ/T), where the exponent ψ = df in the AG theory and ψ = df − θ in the RFOT theory to recover the VFT form, and df is the fractal dimension of the CRR (cooperatively rearranging regions) or “mosaic droplet” for each scenario.59 It has been proposed that the domain size affected by confinement by a free surface or fixed wall might be controlled by the cooperatively rearranging region (CRR) in the AG scenario39 or the mosaic length scale. To extract the size of the mobile surface layer ξ, we fit the local relaxation time profile to an error function log τ(d) = c1 + c2 erf(d/ξ) (see Figure 10b), which we find gives better results for our data than the tanh function39 or exponential form60 used previously by other groups. From the inset of Figure 10c, it is evident that the length scale ξ increases moderately with decreasing temperature, consistent with simulation results on bead−spring chains.13,39 Figure 10c displays a semilog plot of the bulk relaxation time τ as a function of inverse temperature 1/T, normalized by their values at T0 = 500 K. The non-Arrhenius behavior shown in the figure is typical for fragile glass-formers, indicating an increasing activation energy barrier. We plot the same data but as a function of ξ(T)/T and find a reasonably good linear relationship between the logarithm of the bulk relaxation time log(τ) and ξ(T)/T. A similar linear relationship between the logarithm of bulk relaxation time and interfacial length scale normalized by temperature has been previously observed for bead−spring chain models of free-standing49 and supported61 films. This suggests a fractal dimension df ≈ 1 in the AG theory H

DOI: 10.1021/acs.macromol.7b00921 Macromolecules XXXX, XXX, XXX−XXX

Article

Macromolecules m=

∂ log τ ∂(Tg /T )

Tg

(3)

based on the same fit used to determine Tg. For the bulk, we ≈ 344 K and mbulk ≈ 150, which are in obtained Tbulk g reasonable agreement with experimental values65 given that a large extrapolation is used. Remarkably, the depth dependence of local Tg obtained from slow dynamics is consistent with that obtained from various fast dynamical properties, as compared in Figure 10c, which further validates the superposition approach to extract Tg shifts from short-time dynamics. Moreover, a suppression in local fragility is observed near the free surface (see inset of Figure 10d), in line with simulations of bead− spring chains.13 This fragility reduction does not necessarily contradict the collapse of short-time dynamics, since the former may not be reflected in time range probed by the latter.

Figure 11. Example of the shear modes developed in a 22 nm film. Colors are based on the chain positions in the z direction in the initial frame.

■

CONCLUSIONS

In this work, we perform extensive atomistic MD simulations on free-standing polystyrene films to investigate the effect of free surfaces on static and dynamical properties and hence on the glass transition temperature Tg. We modify the TraPPE-UA potential to match the chain backbone dihedral distribution obtained from all-atom simulations, which leads to simulated bulk density and segmental dynamics close to the experimental values. Care has been taken in simulations to ensure consistent results between films and bulk. Despite the high cooling rate and short time scales probed, we find substantial suppression of the dilatometric Tg in thin films inferred from the average film density versus temperature. The dilatometric Tg reduction is much less when we measure only the density in the interior of the film, suggesting that the mobile surface layer is largely responsible for the Tg reduction obtained from the average film density. We also extract both average and local Tg values from dynamical properties. We find the temperature dependence of short-time dynamics, including monomer MSD, chain MSD, and bond MSAD, collapses onto a master curve for different films and the bulk by shifting the temperature for each film thickness. The Tg shifts obtained in this way from short-time dynamics averaged over the whole film exhibits similar film thickness dependence as the dilatometric Tg from average density as well as extrapolated Tg from long-time segmental relaxation. Following the same procedure, we study the local dynamics of each layer in the film at short times and obtain the associated local Tg shift. The local Tg reduction is as large as 80 K near the free surface, vanishes at a distance about 4 nm from the free surface, and is the same for all dynamical properties we investigated. We also show a qualitative agreement between the mobile surface layer thickness based on local Tg inferred from shorttime dynamics and from the range ξ of depth dependence of backbone orientational relaxation times. The latter-defined dynamical length ξ increases moderately with decreasing T and scales with the bulk relaxation time as log(τ) ∼ ξ/T, broadly consistent with both the AG and RFOT scenarios. The discrepancy between computational and experimental results regarding confinement effects on Tg at simulated cooling rate has been one of the most perplexing issues in this field. The main finding in this workthat short-time measurements of static and dynamical properties consistently reveal appreciable

Figure 12. Lateral MSD of monomers at t* = 200 ps as a function of z position before and after the removal the layer center of mass motion. The MSD of center of mass of each layer is well fit to a quadratic function, indicating a linear velocity profile across the z direction. Dashed line is the bulk value with layer center of mass motion removed, which agrees with the value at the center of the film.

Figure 13. Effect of box area on the lateral MSD of layer center of mass and the corrected lateral MSD of monomer at t* = 200 ps.

corresponding to stringlike clusters, or ψ ≈ 1 in the RFOT scenario, consistent with values obtained for different glassformers using different methods.59,62−64 Clearly, data at lower temperature regime and with better statistics are needed before we can draw a firm conclusion as to the value of exponent, which is unfortunately challenging for atomistic simulations. Finally, we estimate Tg from the slow segmental relaxation time τ by extrapolating the VFT fit, τ(T) = τ0 exp(DT0/(T − T0)), to 100 s, as conventionally defined in experiments. In addition, we calculate the fragility index I

DOI: 10.1021/acs.macromol.7b00921 Macromolecules XXXX, XXX, XXX−XXX

Article

Macromolecules

Figure 14. Comparison between different force fields for PS. (a) Probability distributions of dihedral angles along the backbone CHx−CH−CH2− CHy. (b) Probability distributions of improper dihedral angles around the tetrahedron carbon CHx−CH−Caro−CHy. (c) Total pair distribution functions of CH2−CH2 groups. (d) Temperature dependence of bulk densities at P = 1.0 bar with dispersion correction applied. Also shown are experimental data from Zoller and Walsh.69

Figure 12. For higher shear modes, the velocity gradient is larger and the mode is less likely to develop because of the larger shear stress. However, it is difficult to theoretically predict the magnitudes of different modes and their contributions to the apparent shearlike motion, in order to remove them analytically. To correct this artifact, we divided the system into layers about 0.7 nm thick and removed the center-of-mass motion of each layer before analyzing the local dynamics. Figure 12 compares the raw monomer MSD, corrected monomer MSD, and layer center-of-mass MSD as a function of z position for a 22 nm thin film at 500 K. Finally, we verified that this shearlike motion is suppressed by a system with a larger transverse area, while the corrected MSD profile is not affected by the transverse area, as shown in Figure 13 for a 10 nm thick film with different box areas. To keep computational time manageable, rather than increasing the transverse area for thick films, we fix the transverse area of the system and use this technique to eliminate the spurious collective shearlike motions.

Table 1. Modified Dihedral Potential for Atactic Polystyrene V(ϕ) = c1[1 + cos(ϕ)] + c2[1 − cos(2ϕ)] + c3[1 + cos(3ϕ)] + c4[1 − cos(4ϕ)] c1 (kJ/mol)

c2 (kJ/mol)

c3 (kJ/mol)

c4 (kJ/mol)

improper dihedral

5.77 0 c1 (kJ/mol)

−1.23 −2.09 c2 (kJ/mol)

11.33 0 c3 (kJ/mol)

1.01 0 c4 (kJ/mol)

CHx−CH−Caro−CHy

0

3.28

−14.73

−3.36

proper dihedral CHx−CH−CH2−CHy CHx−CH−Caro−CHaro

free surface effects on both average and local Tg in a chemically detailed thin film modelmay shed light on this issue.

■

APPENDIX A. CORRECTION FOR SPURIOUS SHEAR MODES Because of the relatively small box area and correspondingly large fluctuation of the center-of-mass position of each layer, a simulated “tall” thin film (“tall” = transverse dimensions small relative to film thickness) tends to develop shearlike motion along the z direction. This artifact cannot be completely eliminated by removal of the center-of-mass motion of the entire system. In Figure 11 we show two snapshots of a 22 nm film separated by Δt = 16 ns. In addition to the self-diffusion of polymer segments in each layer, shearlike motions develop across the film. This collective motion can be a combination of different modes that satisfy a zero total center of mass motion of the system. For example, the lowest mode is a linear lateral velocity profile along the z-axis, which leads to a quadratic MSD profile for the layer center of mass as a function of height, as shown in

■

APPENDIX B. FORCE FIELD MODIFICATION AND VALIDATION The force field used in this work is adapted from the TraPPEUA model for polystyrene,66,67 modified to match the backbone dihedral distribution obtained from all-atom simulations. We also find it necessary to modify the improper dihedral that maintains the tetrahedral arrangement around the carbon connected to the phenyl ring and to add a torsional potential along the bond joining the phenyl ring to the backbone following previous works36,68 in order to match all-atom results. J

DOI: 10.1021/acs.macromol.7b00921 Macromolecules XXXX, XXX, XXX−XXX

Article

Macromolecules

(10) Ediger, M. D.; Forrest, J. A. Dynamics near Free Surfaces and the Glass Transition in Thin Polymer Films: A View to the Future. Macromolecules 2014, 47, 471−478. (11) Hudzinskyy, D.; Lyulin, A. V.; Baljon, A. R. C.; Balabaev, N. K.; Michels, M. A. J. Effects of Strong Confinement on the GlassTransition Temperature in Simulated Atactic Polystyrene Films. Macromolecules 2011, 44, 2299−2310. (12) Shavit, A.; Riggleman, R. A. Influence of Backbone Rigidity on Nanoscale Confinement Effects in Model Glass-Forming Polymers. Macromolecules 2013, 46, 5044−5052. (13) Hanakata, P. Z.; Douglas, J. F.; Starr, F. W. Local variation of fragility and glass transition temperature of ultra-thin supported polymer films. J. Chem. Phys. 2012, 137, 244901−8. (14) Sussman, D. M.; Schoenholz, S. S.; Cubuk, E. D.; Liu, A. J. Disconnecting structure and dynamics in glassy thin films. arXiv:1310.4876, 2016. (15) Hsu, D. D.; Xia, W.; Song, J.; Keten, S. Glass-Transition and Side-Chain Dynamics in Thin Films: Explaining Dissimilar Free Surface Effects for Polystyrene vs Poly(methyl methacrylate). ACS Macro Lett. 2016, 5, 481−486. (16) Rissanou, A. N.; Harmandaris, V. A. Structural and Dynamical Properties of Polystyrene Thin Films Supported by Multiple Graphene Layers. Macromolecules 2015, 48, 2761−2772. (17) Baljon, A. R. C.; Williams, S.; Balabaev, N. K.; Paans, F.; Hudzinskyy, D.; Lyulin, A. V. Simulated glass transition in freestanding thin polystyrene films. J. Polym. Sci., Part B: Polym. Phys. 2010, 48, 1160−1167. (18) DeFelice, J.; Milner, S. T.; Lipson, J. E. G. Simulating Local Tg Reporting Layers in Glassy Thin Films. Macromolecules 2016, 49, 1822−1833. (19) Forrest, J. A.; Dalnoki-Veress, K. When Does a Glass Transition Temperature Not Signify a Glass Transition? ACS Macro Lett. 2014, 3, 310−314. (20) Priestley, R. D.; Ellison, C. J.; Broadbelt, L. J.; Torkelson, J. M. Structural Relaxation of Polymer Glasses at Surfaces, Interfaces, and In Between. Science 2005, 309, 456−459. (21) Buchenau, U.; Zorn, R. A Relation Between Fast and Slow Motions in Glassy and Liquid Selenium. Europhys. Lett. 1992, 18, 523−528. (22) Simmons, D. S.; Cicerone, M. T.; Zhong, Q.; Tyagi, M.; Douglas, J. F. Generalized localization model of relaxation in glassforming liquids. Soft Matter 2012, 8, 11455−8. (23) Harrowell, P.; Widmer-Cooper, A. Predicting the Long-Time Dynamic Heterogeneity in a Supercooled Liquid on the Basis of ShortTime Heterogeneities. Phys. Rev. Lett. 2006, 96, 185701. (24) Inoue, R.; Kanaya, T.; Nishida, K.; Tsukushi, I.; Telling, M. T. F.; Gabrys, B. J.; Tyagi, M.; Soles, C.; Wu, W. L. Glass transition and molecular mobility in polymer thin films. Phys. Rev. E 2009, 80, 031802. (25) Soles, C. L.; Douglas, J. F.; Wu, W.-L. Dynamics of thin polymer films: Recent insights from incoherent neutron scattering. J. Polym. Sci., Part B: Polym. Phys. 2004, 42, 3218−3234. (26) Ye, C.; Wiener, C. G.; Tyagi, M.; Uhrig, D.; Orski, S. V.; Soles, C. L.; Vogt, B. D.; Simmons, D. S. Understanding the Decreased Segmental Dynamics of Supported Thin Polymer Films Reported by Incoherent Neutron Scattering. Macromolecules 2015, 48, 801−808. (27) Paeng, K.; Swallen, S. F.; Ediger, M. D. Direct Measurement of Molecular Motion in Freestanding Polystyrene Thin Films. J. Am. Chem. Soc. 2011, 133, 8444−8447. (28) Qi, D.; Ilton, M.; Forrest, J. A. Measuring surface and bulk relaxation in glassy polymers. Eur. Phys. J. E: Soft Matter Biol. Phys. 2011, 34, 56−7. (29) Abraham, M. J.; Murtola, T.; Schulz, R.; Páll, S.; Smith, J. C.; Hess, B.; Lindahl, E. GROMACS: High performance molecular simulations through multi-level parallelism from laptops to supercomputers. SoftwareX 2015, 1−2, 19−25. (30) Ramos, J.; Vega, J. F.; Martínez-Salazar, J. Molecular Dynamics Simulations for the Description of Experimental Molecular Con-

In Table 1 we present the modifications to the TraPPE-UA dihedral potentials. All other bonded and nonbonded potential parameters are the same as in ref 66. A comparison of dihedral distributions, pair distribution functions, and densities among different force fields is shown in Figure 14. The modified TraPPE potential used in this work improves the agreement with all-atom results for structural properties. For the bulk density, both the original TraPPE potential and the modified one show good agreement with experimental data.

■

ASSOCIATED CONTENT

* Supporting Information S

The Supporting Information is available free of charge on the ACS Publications website at DOI: 10.1021/acs.macromol.7b00921. Depth dependence of KWW stretching exponent β and autocorrelation function of bond order parameter P2(t) at different temperatures (PDF)

■

AUTHOR INFORMATION

Corresponding Author

*E-mail: [email protected] (S.T.M.). ORCID

Yuxing Zhou: 0000-0002-4476-0680 Scott T. Milner: 0000-0002-9774-3307 Notes

The authors declare no competing financial interest.

■

ACKNOWLEDGMENTS The authors acknowledge stimulating conversations with Mark Ediger, Zahra Fakhraai, Jane Lipson, and Anastasia Rissanou and financial support from NSF DMR-1057980.

■

REFERENCES

(1) Keddie, J. L.; Jones, R.; Cory, R. A. Size-dependent depression of the glass transition temperature in polymer films. Europhys. Lett. 1994, 27, 59. (2) Tress, M.; Erber, M.; Mapesa, E. U.; Huth, H.; Müller, J.; Serghei, A.; Schick, C.; Eichhorn, K.-J.; Voit, B.; Kremer, F. Glassy Dynamics and Glass Transition in Nanometric Thin Layers of Polystyrene. Macromolecules 2010, 43, 9937−9944. (3) Serghei, A.; Huth, H.; Schick, C.; Kremer, F. Glassy Dynamics in Thin Polymer Layers Having a Free Upper Interface. Macromolecules 2008, 41, 3636−3639. (4) Green, P. F.; Glynos, E.; Frieberg, B. Polymer films of nanoscale thickness: linear chain and star-shaped macromolecular architectures. MRS Commun. 2015, 5, 423−434. (5) Fakhraai, Z.; Forrest, J. A. Probing Slow Dynamics in Supported Thin Polymer Films. Phys. Rev. Lett. 2005, 95, 025701. (6) Gao, S.; Koh, Y. P.; Simon, S. L. Calorimetric Glass Transition of Single Polystyrene Ultrathin Films. Macromolecules 2013, 46, 562− 570. (7) Olson, E. A.; Zhang, M.; Zhang, Z.; Allen, L. H. Probing Glass Transition of Ultrathin Polymer Films at a Time Scale of Seconds Using Fast Differential Scanning Calorimetry. Macromolecules 2004, 37, 4607−4616. (8) Huth, H.; Minakov, A. A.; Schick, C. Differential AC-chip calorimeter for glass transition measurements in ultrathin films. J. Polym. Sci., Part B: Polym. Phys. 2006, 44, 2996−3005. (9) Lupaşcu, V.; Picken, S. J.; Wübbenhorst, M. Cooperative and non-cooperative dynamics in ultra-thin films of polystyrene studied by dielectric spectroscopy and capacitive dilatometry. J. Non-Cryst. Solids 2006, 352, 5594−5600. K

DOI: 10.1021/acs.macromol.7b00921 Macromolecules XXXX, XXX, XXX−XXX

Article

Macromolecules

(51) Morita, H.; Tanaka, K.; Kajiyama, T.; Nishi, T.; Doi, M. Study of the Glass Transition Temperature of Polymer Surface by CoarseGrained Molecular Dynamics Simulation. Macromolecules 2006, 39, 6233−6237. (52) Forrest, J. A. What can we learn about a dynamical length scale in glasses from measurements of surface mobility? J. Chem. Phys. 2013, 139, 084702−4. (53) Lipson, J. E. G.; Milner, S. T. Local and Average Glass Transitions in Polymer Thin Films. Macromolecules 2010, 43, 9874− 9880. (54) Mirigian, S.; Schweizer, K. S. Communication: Slow relaxation, spatial mobility gradients, and vitrification in confined films. J. Chem. Phys. 2014, 141, 161103−6. (55) Mangalara, J. H.; Mackura, M. E.; Marvin, M. D.; Simmons, D. S. The relationship between dynamic and pseudo-thermodynamic measures of the glass transition temperature in nanostructured materials. J. Chem. Phys. 2017, 146, 203316−9. (56) Adam, G.; Gibbs, J. H. On the temperature dependence of cooperative relaxation properties in glass-forming liquids. J. Chem. Phys. 1965, 43, 139−146. (57) Kirkpatrick, T. R.; Thirumalai, D.; Wolynes, P. G. Scaling concepts for the dynamics of viscous liquids near an ideal glassy state. Phys. Rev. A: At., Mol., Opt. Phys. 1989, 40, 1045. (58) Bouchaud, J.-P.; Biroli, G. On the Adam-Gibbs-KirkpatrickThirumalai-Wolynes scenario for the viscosity increase in glasses. J. Chem. Phys. 2004, 121, 7347. (59) Starr, F. W.; Douglas, J. F.; Sastry, S. The relationship of dynamical heterogeneity to the Adam-Gibbs and random first-order transition theories of glass formation. J. Chem. Phys. 2013, 138, 12A541. (60) Kob, W.; Roldán-Vargas, S.; Berthier, L. Non-monotonic temperature evolution of dynamic correlations in glass-forming liquids. Nat. Phys. 2011, 8, 164−167. (61) Hanakata, P. Z.; Douglas, J. F.; Starr, F. W. Interfacial mobility scale determines the scale of collective motion and relaxation rate in polymer films. Nat. Commun. 2014, 5, R461. (62) Cammarota, C.; Cavagna, A.; Gradenigo, G.; Grigera, T. S.; Verrocchio, P. Numerical determination of the exponents controlling the relationship between time, length, and temperature in glassforming liquids. J. Chem. Phys. 2009, 131, 194901. (63) Hocky, G. M.; Markland, T. E.; Reichman, D. R. Growing Pointto-Set Length Scale Correlates with Growing Relaxation Times in Model Supercooled Liquids. Phys. Rev. Lett. 2012, 108, 225506. (64) Zhou, Y.; Milner, S. T. Static lengths in glass-forming monodisperse hard-sphere fluids from periodic array pinning. Soft Matter 2016, 12, 402−407. (65) Novikov, V. N.; Ding, Y.; Sokolov, A. P. Correlation of fragility of supercooled liquids with elastic properties of glasses. Phys. Rev. E 2005, 71, 061501−12. (66) Harmandaris, V. A.; Adhikari, N. P.; van der Vegt, N. F. A.; Kremer, K. Hierarchical Modeling of Polystyrene: From Atomistic to Coarse-Grained Simulations. Macromolecules 2006, 39, 6708−6719. (67) Wick, C. D.; Martin, M. G.; Siepmann, J. I. Transferable Potentials for Phase Equilibria. 4. United-Atom Description of Linear and Branched Alkenes and Alkylbenzenes. J. Phys. Chem. B 2000, 104, 8008−8016. (68) Mondello, M.; Yang, H.-J.; Furuya, H.; Roe, R.-J. Molecular Dynamics Simulation of Atactic Polystyrene. 1. Comparison with Xray Scattering Data. Macromolecules 1994, 27, 3566−3574. (69) Zoller, P.; Walsh, D. J. Standard Pressure-Volume-Temperature Data for Polymers; Technomic Pub Co.: Lancaster, PA, 1995.

formation, Melt Dynamics, and Phase Transitions in Polyethylene. Macromolecules 2015, 48, 5016−5027. (31) Yamamoto, T. Molecular Dynamics of Crystallization in a Helical Polymer Isotactic Polypropylene from the Oriented Amorphous State. Macromolecules 2014, 47, 3192−3202. (32) Müller-Plathe, F. Local structure and dynamics in solventswollen polymers. Macromolecules 1996, 29, 4782−4791. (33) Harmandaris, V. A.; Kremer, K. Dynamics of Polystyrene Melts through Hierarchical Multiscale Simulations. Macromolecules 2009, 42, 791−802. (34) Harmandaris, V. A.; Floudas, G.; Kremer, K. Temperature and Pressure Dependence of Polystyrene Dynamics through Molecular Dynamics Simulations and Experiments. Macromolecules 2011, 44, 393−402. (35) Lyulin, A. V.; Michels, M. A. J. Molecular Dynamics Simulation of Bulk Atactic Polystyrene in the Vicinity of Tg. Macromolecules 2002, 35, 1463−1472. (36) Vorselaars, B.; Lyulin, A. V.; Michels, M. A. J. Development of Heterogeneity near the Glass Transition: Phenyl-Ring-Flip Motions in Polystyrene. Macromolecules 2007, 40, 6001−6011. (37) Mansfield, K. F.; Theodorou, D. N. Molecular dynamics simulation of a glassy polymer surface. Macromolecules 1991, 24, 6283−6294. (38) Daoulas, K. C.; Harmandaris, V. A.; Mavrantzas, V. G. Detailed Atomistic Simulation of a Polymer Melt/Solid Interface: Structure, Density, and Conformation of a Thin Film of Polyethylene Melt Adsorbed on Graphite. Macromolecules 2005, 38, 5780−5795. (39) Shavit, A.; Riggleman, R. A. Physical Aging, the Local Dynamics of Glass-Forming Polymers under Nanoscale Confinement. J. Phys. Chem. B 2014, 118, 9096−9103. (40) Dalnoki-Veress, K.; Forrest, J. A.; Murray, C.; Gigault, C.; Dutcher, J. R. Molecular weight dependence of reductions in the glass transition temperature of thin, freely standing polymer films. Phys. Rev. E: Stat. Phys., Plasmas, Fluids, Relat. Interdiscip. Top. 2001, 63, 031801− 10. (41) Pye, J. E.; Roth, C. B. Two Simultaneous Mechanisms Causing Glass Transition Temperature Reductions in High Molecular Weight Freestanding Polymer Films as Measured by Transmission Ellipsometry. Phys. Rev. Lett. 2011, 107, 235701−5. (42) Forrest, J. A.; Dalnoki-Veress, K.; Stevens, J. R.; Dutcher, J. R. Effect of free surfaces on the glass transition temperature of thin polymer films. Phys. Rev. Lett. 1996, 77, 2002−2005. (43) Buenviaje, C.; Dinelli, F.; Overney, R. M. Glass Transition Measurements of Ultrathin Polystyrene Films. ACS Symp. Ser. 2000, 781, 76−92. (44) Schawe, J. E. K. Measurement of the thermal glass transition of polystyrene in a cooling rate range of more than six decades. Thermochim. Acta 2015, 603, 128−134. (45) Mattsson, J.; Forrest, J. A.; Börjesson, L. Quantifying glass transition behavior in ultrathin free-standing polymer films. Phys. Rev. E: Stat. Phys., Plasmas, Fluids, Relat. Interdiscip. Top. 2000, 62, 5187− 5200. (46) Lyulin, A. V.; Balabaev, N. K.; Michels, M. A. J. MolecularWeight and Cooling-Rate Dependence of Simulated Tgfor Amorphous Polystyrene. Macromolecules 2003, 36, 8574−8575. (47) Li, Z.; Yuan, F.; Fichthorn, K. A.; Milner, S. T.; Larson, R. G. Molecular View of Polymer/Water Interfaces in Latex Paint. Macromolecules 2014, 47, 6441−6452. (48) Tsige, M.; Taylor, P. L. Simulation study of the glass transition temperature in poly(methyl methacrylate). Phys. Rev. E: Stat. Phys., Plasmas, Fluids, Relat. Interdiscip. Top. 2002, 65, 021805−8. (49) Lang, R. J.; Simmons, D. S. Interfacial Dynamic Length Scales in the Glass Transition of a Model Freestanding Polymer Film and Their Connection to Cooperative Motion. Macromolecules 2013, 46, 9818− 9825. (50) Varnik, F.; Baschnagel, J.; Binder, K. Reduction of the glass transition temperature in polymer films: A molecular-dynamics study. Phys. Rev. E: Stat. Phys., Plasmas, Fluids, Relat. Interdiscip. Top. 2002, 65, 021507−14. L

DOI: 10.1021/acs.macromol.7b00921 Macromolecules XXXX, XXX, XXX−XXX